Abstract:Manually labeling neurons from high-resolution but noisy and low-contrast optical microscopy (OM) images is tedious. As a result, the lack of annotated data poses a key challenge when applying deep learning techniques for reconstructing neurons from noisy and low-contrast OM images. While traditional tracing methods provide a possible way to efficiently generate labels for supervised network training, the generated pseudo-labels contain many noisy and incorrect labels, which lead to severe performance degradation. On the other hand, the publicly available dataset, BigNeuron, provides a large number of single 3D neurons that are reconstructed using various imaging paradigms and tracing methods. Though the raw OM images are not fully available for these neurons, they convey essential morphological priors for complex 3D neuron structures. In this paper, we propose a new approach to exploit morphological priors from neurons that have been reconstructed for training a deep neural network to extract neuron signals from OM images. We integrate a deep segmentation network in a generative adversarial network (GAN), expecting the segmentation network to be weakly supervised by pseudo-labels at the pixel level while utilizing the supervision of previously reconstructed neurons at the morphology level. In our morphological-prior-guided neuron reconstruction GAN, named MP-NRGAN, the segmentation network extracts neuron signals from raw images, and the discriminator network encourages the extracted neurons to follow the morphology distribution of reconstructed neurons. Comprehensive experiments on the public VISoR-40 dataset and BigNeuron dataset demonstrate that our proposed MP-NRGAN outperforms state-of-the-art approaches with less training effort.

What problem does this paper attempt to address?

The paper attempts to address the issue of the lack of accurately annotated data when reconstructing neurons in optical microscope (OM) images. Specifically: 1. **Manual annotation of neurons is difficult**: Manually annotating neurons from high-resolution but noisy and low-contrast optical microscope images is a tedious task, leading to a severe shortage of annotated data. 2. **Limitations of traditional methods**: Traditional tracing methods can generate pseudo-labels for supervised network training, but these pseudo-labels contain a lot of noise and incorrect labels, resulting in significant performance degradation. 3. **Utilization of existing datasets**: Although the publicly available dataset BigNeuron provides a large number of individual 3D neurons reconstructed through different imaging paradigms and tracing methods, the original OM images in these datasets are not fully available. However, these reconstructed neurons convey the basic morphological prior information of complex 3D neuron structures. To address these issues, the paper proposes a new method that leverages morphological prior information from reconstructed neurons to train deep neural networks to extract neuron signals from OM images. The specific methods include: - **Weakly supervised neuron reconstruction**: By integrating a deep segmentation network and a generative adversarial network (GAN), the segmentation network is expected to be weakly supervised at the pixel level by pseudo-labels while utilizing the morphological supervision of reconstructed neurons. - **Morphological prior-guided GAN**: In the proposed MP-NRGAN framework, the segmentation network extracts neuron signals from the original images, while the discriminator network encourages the extracted neurons to follow the morphological distribution of the reconstructed neurons. Experimental results show that this method outperforms existing state-of-the-art methods on the public datasets VISoR-40 and BigNeuron, and requires less training effort.

Weakly Supervised Neuron Reconstruction From Optical Microscopy Images With Morphological Priors

Neuronal Population Reconstruction From Ultra-Scale Optical Microscopy Images via Progressive Learning

Progressive Learning for Neuronal Population Reconstruction from Optical Microscopy Images

Fast and Lightweight Network Improves Serial Brain Section Stitching

Using Simulated Training Data of Voxel-Level Generative Models to Improve 3D Neuron Reconstruction

Neuron Image Segmentation via Learning Deep Features and Enhancing Weak Neuronal Structures

Semi-Supervised Neuron Segmentation via Reinforced Consistency Learning

A General Deep Learning framework for Neuron Instance Segmentation based on Efficient UNet and Morphological Post-processing

NRTR: Neuron Reconstruction with Transformer from 3D Optical Microscopy Images

Optimization Algorithms in Reconstructions of Neuron Morphology: an Overview

Neuron Segmentation Based on CNN with Semi-Supervised Regularization.

DINeuro: Distilling Knowledge from 2D Natural Images via Deformable Tubular Transferring Strategy for 3D Neuron Reconstruction

Semi-Automatic Neuron Tracing in Electron Microscope Images: Preliminary Results

Two-stage Generative Models of Simulating Training Data at the Voxel Level for Large-Scale Microscopy Bioimage Segmentation

Learning meaningful representation of single-neuron morphology via large-scale pre-training

Attention based morphological guided deep learning network for neuron segmentation in electron microscopy

A Neurocomputational Method for Fully Automated 3D Dendritic Spine Detection and Segmentation of Medium-Sized Spiny Neurons

BigNeuron: a resource to benchmark and predict performance of algorithms for automated tracing of neurons in light microscopy datasets

Learning Multimodal Volumetric Features for Large-Scale Neuron Tracing

DNeuroMAT: A Deep-Learning-Based Neuron Morphology Analysis Toolbox

Brain Image Segmentation for Ultrascale Neuron Reconstruction via an Adaptive Dual-Task Learning Network